Beyond detuning: 10 years of progress and new challenges in the development and application of assays for HIV incidence estimation

Introduction In the ongoing battle against HIV/AIDS, it is critical that we are able to measure and monitor HIV incidence, that is the number of new infections during a period of time, usually expressed as number of infections/person-years of observation or as an annual percentage of the population that acquire infection. Knowledge of HIV incidence is necessary to understand transmission patterns; to provide a rational basis for targeting prevention efforts; to evaluate interventions to reduce transmission; and to predict or project the burden of HIV infection in different demographic and risk populations. Reliable information on HIV incidence is especially important to support prevention programs in the low-income and middle-income countries that continue to bear a disproportionate share of the global burden of the HIV epidemic. Improved estimates of HIV incidence are essential to evaluate ongoing HIV prevention and treatment programs in these resource-constrained settings and to guide the most effective use of the billions of dollars that will be spent on the epidemic in the coming years [1]. Direct measurement of incidence through prospective follow-up of cohorts of HIV-negative persons is expensive, generally unrepresentative of the larger population, and difficult to sustain, even in resource-rich settings. Furthermore, enrollment of persons into cohort studies can introduce selection bias or induce behavior change that can result in a lower observed HIV incidence than is representative of the population from which the cohort is drawn. As a consequence of the shortcomings of longitudinal studies, HIV incidence estimates have frequently been provided by mathematical models based on epidemiological information about HIV prevalence and AIDS diagnoses or death rates. When input parameters are well established, these model-based approaches can provide reasonably accurate indirect incidence estimates [2]. Most recently, an epidemiological model for incidence estimation has been described that compares prevalence data in two sequential cross-sectional prevalence surveys [3]. This method has been validated using simulated data and through comparison of its estimates with measurements of incidence in several community-based cohort studies. Measuring the impact of antiretroviral treatment on age-specific HIV prevalence levels has further improved the epidemiological HIV incidence estimation from survey data [4]. However, the application of this method to estimate HIV incidence is limited to the general population at the national level, because there is a fundamental assumption about individuals in different age groups being comparable in the surveys. This situation would not be the case for prevalence data obtained from subpopulations in which individuals may enter and exit as they age (such as injecting drug-users or sex workers). Thus, classic and newer modelling approaches to HIV incidence estimation are logistically challenging, are difficult to standardize over time, and/or require complicated statistical methods. Given the complexities and limitations of epidemiological and modeling approaches to measure HIV incidence, there has long been a strong argument for laboratory methods that can distinguish recent from established or long-term HIV infection, in order to estimate HIV incidence. In this Review, we present a summary of past and present efforts to develop and apply HIV incidence assays, and discuss what it will take to develop improved HIV incidence assays and algorithms, and to support accurate application of these methods in various research and surveillance settings. Insights that led to development of cross-sectional HIV incidence strategies In the early 1990s, our understanding of the dynamics of early HIV infection began to improve. The concepts of ‘incidence-window period’ (I-WP) modeling were first framed out by Petersen et al.[5] at the Centers for Disease Control and Prevention (CDC) and collaborators in the United States blood banking community. To estimate the window period preceding development of HIV antibodies, these investigators used data from recipients of blood transfusions who became infected with HIV, i.e., so-called ‘look-back’ investigations (i.e., tracing and follow-up testing of transfusion recipients who received blood from donors who subsequently seroconverted). The infectious window period was estimated to last 56 days prior to HIV seroconversion based on first-generation antibody assays and 45 days prior to seroconversion for second-generation HIV antibody assays. When this information was combined with longitudinal cohort data on HIV incidence in repeat blood donors, the residual risk of transmitting HIV from preseroconversion ‘window-phase donations’ could be calculated (residual risk = I × WP) [6]. Subsequent detailed studies of viral nucleic acid, p24 antigen and antibody dynamics in acute HIV infection led to refined understanding of the evolution of laboratory markers during the first several months of HIV infection [7,8]. For example, Fiebig et al.[8] tested panels of specimens from seroconverters against an array of readily available laboratory assays for HIV viremia and antibodies. They were able to generate precise point estimates with confidence intervals (CIs) for serial stages of acute infection, as defined by the results of these multiple HIV tests. Fiebig observed the ability of different HIV assays to ‘close the preseroconversion window period,’ and projected how the use of different tests might affect the yield of testing in populations with different rates of HIV incidence (Yield = I × ΔWP). This approach was widely embraced by test manufacturers, regulators and policy makers, who shared the common goal of closing the preseroconversion window period to enhance blood safety. The window period concept was first inverted to estimate HIV incidence by Brookmeyer and Quinn [9]. They developed an approach for testing seronegative samples for p24 antigen and used the rate of detection of these incident infections (i.e., those in the preantibody seroconversion, antigenemic phase) and the known duration of the preseroconversion window period, to calculate incidence. Although innovative, this approach was problematic owing to the logistics and expense inherent in testing large numbers of seronegative samples for p24 antigen, which turned out to be detectable for only 5–10 days prior to antibody appearance. Moreover, these authors noted that the rate of detection of p24 antigen-positive/antibody-negative samples appeared to overestimate incidence, relative to rates of seroconversion documented on follow-up. They attributed this to test seeking during the acute symptomatic phase of infection (resulting in higher cross-sectional incidence estimates) and/or to modified risk following initial testing and counseling (resulting in lower observed incidence rates). In recent years, a number of HIV testing programs have implemented similar tests for acute HIV infection, using highly sensitive HIV RNA tests to detect viremic/preseroconversion HIV testing clients [10,11], and reviewed in Zetola and Pilcher [12]. This experience has consistently shown that early, preseroconversion HIV infections are indeed greatly overrepresented in high-risk testing settings. Concurrent with Brookmeyer and Quinn's work, Janssen et al.[13] developed and validated a ‘detuned’ or less-sensitive HIV antibody assay, which could discriminate recent from long-standing infections, as a tool to estimate incidence in large cross-sectional populations. In a landmark 1998 article published in JAMA, Janssen et al.[13] presented this new approach to measuring HIV incidence. The technique, subsequently termed the Serological Testing Algorithm for Recent HIV Seroconversion (STARHS), was based on the principle that HIV antibody titers evolve in a predictable fashion over the months following initial seroconversion. Laboratory assays could, thus, be developed that would differentiate recent from long-standing HIV infections in a testing population, based on antibody evolution. The first such assay employed a second generation commercial HIV antibody assay (Abbott HIV 3A11) performed using less-sensitive or ‘detuned’ conditions (high dilution, reduced incubation periods, high cutoff). Specimens from recently infected persons test reactive on the standard antibody assay protocol, but nonreactive on the less-sensitive assay protocol. Based on the analysis of a large number of seroconversion panels and specimens, a series of ‘detuned window periods’ were calculated for different cutoffs of the modified less-sensitive assay. For example, at a standardized optical-density cutoff of 1.0, the window period from seroconversion on the standard assay to seroconversion on the less-sensitive assay was 129 days (95% CI: 109–149 days). The rate of detecting persons in the ‘detuned window period’ (less sensitive-enzyme immunoassay (LS-EIA) WP) could then be used to calculate incidence in the tested population using the simple formula: I = rate of recent seroconvertors/LS-EIA WP/365, analgous to the approach suggested by Brookmeyer and Quinn. A similar approach was later developed for a second licensed HIV-1 antibody assay (the BioMérieux Vironostika EIA) [14,15]. Both assays were widely used to calculate incidence projections for various populations in the late 1990s and early 2000s (reviewed by McDougal et al.[16] and Le Vu et al.[17]). With broad use of the STARHS approach, this testing was found to have a number of potentially serious limitations. Most critically, detuned assays tended to misclassify as ‘recent’ persons with late-stage AIDS (with depleted HIV immunity and hence waning antibodies), as well as persons with viral suppression owing to antiretroviral therapy (ART) in whom antibodies also wane [18]. The performance of STARHS using the earliest detuned assays was also found to vary according to viral subtypes in the tested population. These findings led to the need to exercise caution in selection of populations for application of STARHS – exclusion of AIDS patients and ART-treated patients, and evaluation of subtype-specific window periods for each recent infection assay used to calculate incidence. These limitations led to efforts to develop alternative laboratory methods that might be less susceptible to these confounding factors. Current laboratory methods for detection of recent HIV infection and their limitations Other immune-response maturation approaches to identifying recent HIV-1 infection have been developed over the last 6 years (Fig. 1). These include assays based on characterizing the proportion of total bound immunoglobulin G (IgG) antibodies that capture a synthetic branched gp41 peptide constructed to express immunodominant epitopes of HIV-1 subtypes B, E, and D (the ‘BED’ capture enzyme immunoassay) [19]; quantification of the avidity of anti-HIV antibodies using modified second or third generation anti-HIV assays [20–22]; measurement of the antibody response to a gp41 immunodominant epitiope (IDE) and various gp120-V3 loop peptides (the ‘IDE-V3’ assay) [23–25]; measurement of isotype IgG3 anti-HIV, which is present early in the immune response [26]; and quantification of anti-HIV antibodies on a line immunoassay (Inno-LIA HIV adaptation) [27]. All of these assays have shown promise and many have been employed in screening programs to detect recent seroconverters for clinical, public health, and research purposes, and to measure and track HIV incidence (recently reviewed in [17,28,29]). Several rapid HIV antibody assays have also been adapted for detection of recent seroconversion and incidence estimation [30,31]. The most recent proposal is a new avidity assay developed by CDC in 2009 [32]. Of the assays that are currently available, the BED is the most commonly used. A summary of available assays and their limitations are presented in Table 1.Fig. 1: Principle of assays that discriminate recent from long-standing HIV infections, based on maturation of HIV-specific antibody responses, for use in cross-sectional incidence estimation. LS-EIA, less-sensitive-enzyme immunoassay.Table 1: Summary of tests for recent infection and their limitations.Each of these approaches continues to be challenged by the variability of the immune response among recent HIV-1-infected persons, by differential performance in populations infected with different HIV-1 subtypes, and by the impact of ART, late-stage AIDS immunosuppression, and ‘elite controller’ status on individual subject misclassification. It remains unclear exactly how, or to what extent each of these factors influences the precision and specificity of available serologic assays in determining recent HIV seroconversion. However, such factors can lead to an overall lack of specificity in identifying persons with recent infection. Moreover, there is mounting evidence that many STARHs assays (and particularly the BED assay) are susceptible to high rates of misclassification of individuals with long-standing infections as recently infected [i.e., high ‘false recent rates’ (FRR)], impairing their performance in many settings. To address this issue, several groups have recently proposed using test algorithms employing two or more incidence assays in sequence, as part of a Recent Infection Testing Algorithm (RITA) to reduce the FRR and increase the accuracy of incidence estimations [1]. An example of the application of a RITA based on two different incidence assays in combination with additional clinical information on CD4 cell count and antiretroviral treatment is shown in Fig. 2. This approach is reasonable and merits further evaluation, but may increase the complexity and cost of applying tests for recent infection [33]. The inclusion of CD4 cell counting in the testing algorithm requires whole blood samples, which implies that this information would not be available with dried blood spot specimens typically collected in national population-based surveys. In addition to these problems with assay performance, problems persist with standardization and quality control, and with the cost and continued availability of the commercial assays that have been adapted for incidence testing [28].Fig. 2: Representative testing algorithm incorporating two incidence assays and available clinical data (CD4 count and antiretroviral treatment history) for determination of recent HIV infection status of specimens evaluated for HIV incidence estimation. Assay 1 and assay 2 represent two assays for recent infection, preferably based on different principles for discriminated recent from non-recent infections.The WHO working group on HIV incidence assays A number of leading public health organizations and scientists involved in incidence assay development and application have recently launched a concerted effort to overcome technical barriers, and to develop robust approaches based on immune maturation testing for incidence estimation. To advance this effort, in 2008, WHO formally convened a Technical Working Group on HIV Incidence Assays (http://www.who.int/diagnostics_laboratory/links/hiv_incidence_assay/en/index.html). This group is made up of epidemiologists, laboratory specialists, and public health officials. Several gaps were identified in assay development, validation, and commercialization. For example, there has been a lack of standard terminology and no clear consensus on statistical methods to use for HIV incidence estimation. It was also evident that the assays currently available and employed have not yet been rigorously validated using appropriate, comprehensive sample sets that would allow a standardized comparison of their performance, and a of the of assays into The the need for assays for estimation of HIV incidence to support prevention and has provided and support for the Working Group to these limitations and an initial of the on this the Working Group has standardized for assay development, new for HIV incidence a and of specimens to be used for HIV incidence validation, and to incidence assays. A was also convened to address in statistical methods for cross-sectional HIV incidence estimation based on testing for recent infection. terminology initial development of assays for recent HIV infection in the late 1990s, various have been used to assays and methods to estimate HIV incidence. In order to WHO convened several to discuss the most to assays for recent infection. The results of this are in Table 2 2: terminology and for the application of assays for recent infection to estimate HIV In a global and to estimate for HIV incidence assays with to the for HIV incidence assays, that of acute and recent HIV infection are and are used The BED assay was the most widely known and the most commonly used incidence assay among Although the BED assay was widely and use of this assay were Although its cost and many investigators with its use the BED to overestimate HIV incidence. a for assays with improved performance and were to in of that assay performance would lead to for their and surveillance were as influences on In to HIV that are for clinical for HIV incidence assays is by public health and research projected the for 1: The in which HIV incidence assays are used only on specimens and is by 2: An improved HIV incidence assay is widely available at a cost and consistently well HIV-1 This new assay is used only on A HIV incidence assay is developed that can be used to and to how recently a sample was This assay would be used for public health and research purposes, than individual of each of the described are provided in Table The results that an assay that could be used to a sample as and from a recent infection could result in an of specimens tested over This would be in to the use of such assays to specimens of as to use only on specimens as for HIV incidence assays. However, the limitations of the assays that are currently available have been widely noted and have not their and is number of specimens to be tested for HIV incidence over years, by and new The of on testing for recent HIV infection has on the use of modified serologic assays for detection of HIV nucleic and antibodies. based on have been proposed for measuring incidence that may these may evolve in predictable in to early serologic in acute HIV infection collected to that in are limited to the early period, potentially their for incidence estimation. viral typically in the months following infection and the of the first of antibody simple assays for of HIV can be and in by such assays are more or less accurate than serologic assays, to be in population over it has recently been proposed that it might be to measure HIV incidence by measuring how population viral over methods have recently been developed that might be adapted for incidence estimation et Although is expensive, the that many surveillance programs data that could be used to data for incidence estimation. To of the approaches have been to evaluation with to their application in incidence estimation. For the the WHO Working Group remains on the further development and of approaches based on serologic assays for recent infection. The group will a for or more of these alternative as well as the use of HIV RNA individuals with viremia in and and CD4 cell count data detect AIDS for and exclusion of of long-standing infection that are as recent infection (i.e., reduce the FRR of of specimens available for use in measuring and infection testing algorithm performance An essential of the development and evaluation for the and of assays for recent infection is the availability of sample An ongoing research has been launched to and studies that could potentially specimens to panels that could be used to and evaluate assays for recent HIV infection. of and survey results from this effort are available in a to be made available at were able to of studies of and specimens that would be highly to evaluation of tests for recent infection cohorts and studies known infected patients, and patients with However, the of these research studies specimens in only limited and for limited numbers of In of these the WHO Working Group has an effort to large numbers of well and blood specimens, to allow evaluation of currently available assays. The that is currently being will include specimens, but will also include specimens collected for this by cohorts and blood programs with to of of statistical methods for incidence estimation In a statistical of the WHO Working Group to discuss the in the of cross-sectional incidence the RITA – described as the RITA duration or and the FRR – are known with in principle this incidence to be estimated with different of this method have been described and it has been shown that this approach limitations of approaches it is to calculate the of incidence estimates by into the statistical of in the cross-sectional as well as the with which the performance of the RITA are the methods by the WHO Working Group can in the of sample and and are available on the of the for and A for characterizing a RITA in of a RITA and a FRR also an improved analysis of the fundamental by of tests. In it is to have the duration of the recent to be as long as survey sample a FRR the statistical of for this These two factors are in and this is the of in identifying and The of statistical with an FRR is particularly from which it that estimates of the FRR can lead to bias in incidence estimation. analysis of incidence can be a reasonably precise estimate of a of the order of or two statistical tests for can at be performed the FRR in the tested population. Although the need for laboratory approaches to measure HIV incidence in populations is the use of tests for recent infection is greatly by problems test performance, and by the lack of data to guide the use and of methods. In order to this critical the WHO Working Group on HIV incidence assays that a concerted effort be to rigorously evaluate the performance of laboratory assays for recent to develop for use and for surveillance and research and to develop working public and to the frequently In this the need for recent infection testing is most by the public health which need information on HIV incidence. investigators and of and need to high incidence subpopulations in which to studies and then measure incidence as important of the lack of for development and evaluation of recent infection assays, it is that public and will need to a to advance development and availability and application of for incidence estimation. this effort, over a in is for the of the HIV